Research map / Point clouds / Point cloud classification
Recurrent Slice Networks for 3D Segmentation of Point Clouds
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add slice pooling, bidirectional RNNs, and slice unpooling to model local point dependencies | PointNet-style point-cloud segmentation · S3DIS train/test split | mIOU: 41.09 -> 51.93 | Helped |
| Add slice pooling, bidirectional RNNs, and slice unpooling to model local point dependencies | PointNet++ · ScanNet | mIOU: 34.26 -> 39.35 | Helped |
| Use RSNet with xyz inputs instead of PointNet++ trained without normal information | ShapeNet part segmentation · ShapeNet | mean: 84.6 -> 84.9 | Helped |
| Set slice resolution to 2 cm on all three axes instead of the tested alternatives (ablation) | RSNet · S3DIS train/test split | mIOU: 49.12, 51.93, 51.20, 49.16 | Helped |
| Set horizontal slice resolution to 2 cm rather than 1, 4, or 6 cm (ablation) | RSNet · S3DIS train/test split | mIOU: 49.23, 51.93, 48.97, 47.86 | Helped |
| Use 1 m sliding blocks instead of larger 2 m or 3 m blocks (ablation) | RSNet · S3DIS | mIOU: 51.93; 44.15, 44.59, 43.15; 39.08, 37.77, 37.55, 37.21, 35.25 | Helped |
| Use overlapping test blocks with stride 0.5 m instead of non-overlapping blocks with stride 1.0 m (ablation) | RSNet · S3DIS test | mIOU: 51.93 -> 53.83 | Helped |
| Use overlapping test blocks with stride 0.2 m instead of non-overlapping blocks with stride 1.0 m (ablation) | RSNet · S3DIS test | mIOU: 51.93 -> 52.39 | Helped |
| Replace vanilla RNN units with GRU units (ablation) | RSNet · S3DIS | mIOU: 45.84 -> 51.93 | Helped |
| Replace LSTM units with GRU units (ablation) | RSNet · S3DIS | mIOU: 50.08 -> 51.93 | Helped |
| Add median frequency balancing during training (ablation) | RSNet · S3DIS train/test split | mAcc: 59.42 -> 62.09 | Helped |
| Add median frequency balancing during training (ablation) | RSNet · S3DIS train/test split | mIOU: 51.93 -> 48.68 | Hurt |
Closest papers by meaning
- PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation (CVPR 2017 · similarity 0.88)
- Learning to Segment 3D Point Clouds in 2D Image Space (CVPR 2020 · similarity 0.86)
- PIE-NET: Parametric Inference of Point Cloud Edges (NeurIPS 2020 · similarity 0.85)
- PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies (NeurIPS 2022 · similarity 0.85)
- SGPN: Similarity Group Proposal Network for 3D Point Cloud Instance Segmentation (CVPR 2018 · similarity 0.85)
- PointAttN: You Only Need Attention for Point Cloud Completion (AAAI 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.